Worst-case identification of nonlinear fading memory systems (Q1893013)

Asymptotic identification of a general class of fading memory nonlinear systems in the presence of bounded additive output noise is investigated. The worst-case asymptotic error as a measure of identification accuracy is used. A characterization of asymptotically optimal algorithms and optimal asymptotic errors is given, and optimal persistently exciting inputs minimizing the radius of the worst-case uncertainty set are then specified. Consequently, a convergent algorithm independent of the noise upper bound is provided. The algorithm is based on interpolating measured data with multivariate piecewise-linear spline functions, which are shown to be better suited for identification under presence of bounded noise than e.g. polynomials.